Mixed-Precision Quantization and Parallel Implementation of Multispectral Riemannian Classification for Brain--Machine Interfaces

TL;DR

This paper presents a mixed-precision quantized multispectral Riemannian classifier for brain--machine interfaces, achieving high accuracy and efficiency on low-power microcontrollers for real-time motor imagery classification.

Contribution

It introduces a novel mixed-precision quantization approach for a Riemannian classifier tailored for embedded BMI applications, improving accuracy and efficiency.

Findings

Achieved 75.1% accuracy on 4-class MI task.

Quantized model with only 1% accuracy loss.

Implemented on MCU with 33.39ms processing time and 1.304mJ energy consumption.

Abstract

With Motor-Imagery (MI) Brain--Machine Interfaces (BMIs) we may control machines by merely thinking of performing a motor action. Practical use cases require a wearable solution where the classification of the brain signals is done locally near the sensor using machine learning models embedded on energy-efficient microcontroller units (MCUs), for assured privacy, user comfort, and long-term usage. In this work, we provide practical insights on the accuracy-cost tradeoff for embedded BMI solutions. Our proposed Multispectral Riemannian Classifier reaches 75.1% accuracy on 4-class MI task. We further scale down the model by quantizing it to mixed-precision representations with a minimal accuracy loss of 1%, which is still 3.2% more accurate than the state-of-the-art embedded convolutional neural network. We implement the model on a low-power MCU with parallel processing units taking only 33.39ms and consuming 1.304mJ per classification.